Bayesian Contrast Measures and Clutter Distribution Determinants of Human Target Detection

Area of Science:

• Visual perception research
• Computational neuroscience
• Image analysis

Background:

• Human target detection is influenced by electro-optical factors, target characteristics, and contextual elements.
• Understanding the interplay of these factors is crucial for improving detection systems.
• Previous models often simplify contrast and clutter effects.

Purpose of the Study:

• To investigate human target detection using a Bayesian framework.
• To develop and compare different contrast measures, including a novel Bayesian-based histogram contrast statistic.
• To analyze the impact of clutter on target detection under various conditions.

Main Methods:

• A Bayesian approach was employed to model human target detection.
• Three contrast formulations were developed and compared: mean contrast, perceptual contrast, and Bayesian histogram contrast.
• Detection data was analyzed to assess the correlation between contrast measures and human performance, considering target size and clutter interactions.

Main Results:

• The Bayesian histogram contrast statistic demonstrated a strong correlation with human detection performance, outperforming other measures.
• Clutter effects were minimal for large targets and when clutter was not contiguous with the target.
• Detection performance decreased significantly when targets were contiguous with similar-feature clutter, creating a "clutter camouflage" effect.

Conclusions:

• A Bayesian formulation incorporating contrast histogram statistics and image context provides a robust model for human target detection.
• Human observers appear to adapt their detection criteria based on image set, context, and task demands.
• The developed contrast measure and clutter analysis offer insights for designing more effective visual search and detection systems.